Fixed Order Analysis: Cross section in histograms does not match summary for high precision

Asked by Alexander Ba on 2018-08-27

Dear community,

I use Madgraph v.2.6.3.2 (the same also appears with v.3.0) with the (unchanged) fixed order analysis analysis_td_template.f.
I generate following process:

import model loop_sm-no_b_mass
generate p p > t t~ j [QCD]

generate_events
order=LO
fixed_order=ON

When running generate_events with a precision of req_acc_FO = 0.0005 I get a total cross section of 9.685e+01 (summary.txt), while the corresponding entry in the topdrawer file MADatNLO.top (total rate, central value) is much larger (13.86D+01). With a lower precision like req_acc_FO = 0.05, however, both numbers agree.

Is there any reason for this deviation? Is there any way to solve it, or is it safe to simply rescale the output histograms to the correct cross section?

Thank you very much in advance!

Question information

Language:
English Edit question
Status:
Solved
For:
MadGraph5_aMC@NLO Edit question
Assignee:
Rikkert Frederix Edit question
Solved by:
Alexander Ba
Solved:
2018-09-04
Last query:
2018-09-04
Last reply:
2018-09-03

This question was reopened

Rikkert Frederix (frederix) said : #2

Dear Alexander,

Thanks for pointing this out.
Indeed, there is a serious bug in the code that leads to wrong distributions in the case of high-precision runs (i.e., with a low req_acc_FO) when using the TopDrawer format. (The default "HwU" format nor the "root" format are not affected). It is NOT correct to simply rescale the distributions to the correct cross section.

I've just written a patch. It can be found here

http://users.ph.tum.de/ga63zot/topdrawer.patch.gz

This will become part of the upcoming 2.6.4/3.0.1 releases of the code. Please let me know if this solves your problem!

Best regards,
Rikkert

Alexander Ba (aboson) said : #3

Thanks Rikkert Frederix, that solved my question.

Alexander Ba (aboson) said : #4

Dear Rikkert Frederix,

Thank you very much for answering that fast and for providing the patch. However, I have the feeling that it does not work 100% correctly.

In case I use 16 cores (nbcore=16), the results of the cross section in the summary.txt (98.13 pb) and MADatNLO.top (95.41 pb) are close, but not exactly the same. With 64 cores, however, the results still differ significantly (36.56 pb in MADatNLO.top, 98.13 pb in summary.txt).

With HwU as output, on the other hand, the numbers are exactly the same (98.13 pb) in both cases.

Best regards,
Alexander

Rikkert Frederix (frederix) said : #5

Dear Alexander,

Strange. I'll have another look. I'll let you know.

best,
Rikkert

Rikkert Frederix (frederix) said : #6

Dear Alexander,

I think I've found the problem. Please put this file

http://users.ph.tum.de/ga63zot/combine_plots_FO.sh

into your Template/NLO/SubProcesses/ directory (and, if you don't want to regenerate your ttbar+jet process, the <YOUR_PROCESS>/SubProcesses/ directory) --- i.e., replace the file with the same name that's already there. That should fix the problem.

Best,
Rikkert

Alexander Ba (aboson) said : #7

Dear Rikkert,

now it works fine. Thank you very much!

Best,
Alexander